OpenAI has made the Decisions API generally available. The API is built with GPT-6 Luna and is designed for the many small decisions that applications and agents make during a workflow, such as classifying an input, routing a request, or choosing the next action. Individually these calls are simple, but their latency and cost add up at scale. OpenAI says concentrating Luna’s intelligence on bounded-answer questions lets applications make near real-time decisions from text, images, and structured inputs up to 10x faster than GPT-6 Luna through the Responses API.

The developer workflow is straightforward: define the question and allowed answers, then provide context as text, images, or native JSON. The API returns one of three answer types—probability, choice, or score—which the application can use to determine what happens next. Since the public beta, OpenAI has added a native GPT-Live integration, prompt caching for reused context, and inference residency in the US and EU. Developers can pass native JSON inputs, include images in question instructions, and ask up to 400 questions per request, double the previous limit. Latency has also improved, with additional capacity in several regions in Asia.

Examples show the range of applications. Internal OpenAI teams are using Decisions to reduce response times (dots team), categorize transaction descriptions (Personal Finance), and classify coverage by themes and score whether it reinforces or challenges key narratives (Communications). Customer uses described include attributing AI costs and making language-learning experiences more interactive.

With GPT-Live, the Decisions API can be used in voice applications. OpenAI describes a customer-support agent deciding whether to ask a clarifying question, route a request, or escalate an issue, and an avatar that adjusts its facial expressions based on the user’s speech. The concrete example adds a configuration to session.triggers in an existing GPT-Live session; Live emits a decision.event, and the application reads the selected reaction and animates the character.

Pricing and availability: the API is available on a dedicated Decisions endpoint and costs $0.10 per million input tokens; cached input is $0.01 per million tokens. There are no output-token charges. Cached writes are supported for 30-minute and 12-hour windows; regional processing is available in the US and EU, and Zero Data Retention is available to eligible customers. OpenAI suggests starting with one recurring decision, defining the question, possible answers, and needed context, then testing in the Decisions Playground or following the developer guide.